New Statistical Methods for High Resolution Mapping of Multiple Phenotypes

生物医学领域的数据采集能力在过去十年中大幅提高。 遗传学也许是最引人注目的例子：我们从获得第一个序列 通过大规模、多中心的研究， 依赖于数千个实验对象的DNA序列与此同时，技术进步- 这些数据使我们能够以前所未有的精度和分辨率测量人类表型。在 为了利用这些新的大规模数据集中的信息， 很好地适应了问题的规模。 这项建议的重点是制定重新排序分析的统计方法 数据的目的是确定医学相关表型的遗传基础， 可能是多变量的。事实上，我们的动机是分析中出现的具体需要 研究代谢综合征和双相情感障碍的数据集。在目标1中，我们考虑 在这种情况下，重新测序的动机是为了识别影响基因的遗传变异， 表型的基因组位点，其相关性是以前建立的。在目标2中， 当多个样本同时存在时，为鉴定结果的重现性提供保证的挑战 表型是同时研究的。将实施所制定的所有方法 发布到科学界的软件中。 我们计划开发的统计工具是多种多样的：贝叶斯分层建模和创新， 控制错误发现数量的有效策略。它们都很适合于炭-- 当代数据集的特性，允许在高维空间中搜索稀疏信号 空间.一名博士后学者和一名研究生将为该研究项目做出贡献， 他们将获得的培训是拟议工作的额外好处。

Data acquisition capacity in the biomedical field has increased substantially in the last ten years. Genetics is perhaps the most spectacular example: we have gone from obtaining the first sequence of the human genome through a large-scale, multi-center effort to a multiplicity of studies, each relying on the DNA sequence of thousands of subjects. At the same time, technological advance- ments allow us to measure human phenotypes with unprecedented precision and resolution. In order to harness the information in these new large-scale datasets, novel analytical methods that are well adapted to the scale of the problem are needed. This proposal focuses on developing statistical approaches for the analysis of resequencing data with the goal of identifying the genetic underpinning of medically relevant phenotypes, possibly multivariate. In fact, we are motivated by the concrete needs emerging from the analysis of datasets collected to study metabolic syndrome and Bipolar disorder. In Aim 1, we consider the case where re-sequencing is motivated by the goal of identifying genetic variants that influence phenotypes in genomic loci whose relevance was previously established. In Aim 2 we take on the challenge of providing guarantees on the reproducibility of the identified results when multiple phenotypes are investigated simultaneously. All methodology developed will be implemented in software released to the scientific community. The statistical tools we plan to develop are varied: Bayesian hierarchical modeling and inno- vative strategies to control the number of false discoveries. They are all well-adapted to the char- acteristics of contemporary datasets, allowing for search of sparse signals in high-dimensional spaces. A postdoctoral scholar and a graduate student will contribute to the research program, and the training they will acquire is an additional benefit of the proposed work.